Preserving Uncertainty in Demand Prediction for Autonomous Mobility Services

Inon Peled, Kelvin Lee, Yu Jiang, Justin Dauwels, Francisco Camara Pereira

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review


Unlike traditional bus fleets, autonomous mobility services are naturally amenable to dynamic, demand-responsive adaptation of itinerary. Accurate prediction of demand for such services can thus improve their utilization and decrease
their operational costs. Although demand for transit services is inherently stochastic, models of demand often reduce its distribution to point estimates, thus losing useful information for subsequent decision making. In this paper, we advocate for preserving the full predictive distribution through quantile regression, so that the structure of uncertainty in future demand is preserved. To demonstrate our approach, we present a real-world case study of an autonomous shuttle service in a Danish university campus, for which we have several weeks of crowd movement counts, as reconstructed from campus WiFi records. We devise several types of quantile regression models for demand prediction, analyze their performance, and discuss their applicability to the case study. Our modeling methodology can be extended to autonomous fleets of higher scale, thus promoting sustainable shared mobility.
Original languageEnglish
Title of host publicationProceedings of the IEEE Intelligent Transportation Systems Conference (ITSC) 2019
Publication date2019
Publication statusPublished - 2019
EventITSC 2019 : The 22nd IEEE International Conference on Intelligent Transportation Systems - Conference Venue Cordis Hotel, Auckland, Auckland, New Zealand
Duration: 27 Oct 201930 Oct 2019


ConferenceITSC 2019
LocationConference Venue Cordis Hotel, Auckland
Country/TerritoryNew Zealand
Internet address


Dive into the research topics of 'Preserving Uncertainty in Demand Prediction for Autonomous Mobility Services'. Together they form a unique fingerprint.

Cite this